The AI Index Report 2025 from Stanford HAI was recently released. The authors present a 455-page in-depth analysis of the development of AI. Expert on the application of artificial intelligence Alexey Minakov has identified 12 key insights about AI. Here they are.
The AI Index Report 2025 from Stanford HAI was recently released. The authors present a 455-page in-depth analysis of the development of AI. Expert on the application of artificial intelligence Alexey Minakov has identified 12 key insights about AI. Here they are.
AI performance in tests continues to improve. In 2023, new tests appeared — MMMU, GPQA and SWE-bench — to test the capabilities of the most powerful models. In just one year, the indicators have grown sharply: +18.8%, +48.9% and +67.3%, respectively.
AI is rapidly becoming integrated into everyday life. In 2023, the U.S. Food and Drug Administration approved 223 AI-enabled medical devices, up from six in 2015. On the road, self-driving cars are no longer experimental: Waymo, one of the largest U.S. operators, provides more than 150,000 autonomous rides each week, and China’s Apollo Go robo-taxi serves numerous cities across China.
Businesses are embracing AI in droves, investing record amounts and seeing productivity gains. 78% of organizations are already using AI, up from 55% last year. Research shows that AI is truly improving productivity and helping to address skills shortages.
The US leads in the number of top AI models, but China is closing the gap in quality. US institutions produced 40 significant models in 2024, China 15, Europe just 3. However, the quality gap between US and Chinese models on major tests has almost disappeared. China continues to lead in the number of publications and patents in the field of AI.
The field of responsible AI is evolving. The number of AI-related incidents has increased dramatically. Despite this, most companies do not yet use standardized approaches to risk assessment. New safety and reliability tests have emerged (HELM Safety, AIR-Bench, FACTS), but in practice progress remains slow. Instead, governments have stepped up: the EU, the UN, and the African Union have developed frameworks for transparency and trust in AI.
Optimism about AI is growing. In China (83%), Indonesia (80%), and Thailand (77%), the majority of the population considers AI to be more useful than harmful. In Canada (40%), the United States (39%), and the Netherlands (36%), the figures are much lower. However, the dynamics are positive: since 2022, optimism has increased in Germany, France, Canada, the United Kingdom, and the United States.
AI is becoming more efficient, cheaper, and more accessible. The cost of using models at the GPT-3.5 level has decreased by more than 280 times from November 2022 to October 2024. The cost of hardware is falling by 30% every year, and energy efficiency is increasing by 40%. Open models have almost caught up with closed ones: the performance gap has narrowed from 8% to 1,7% in just one year.
Governments are stepping up to regulate and invest in AI. In the US, federal agencies have enacted 59 new AI regulations, double the number in 2023. Across 75 countries, AI legislation has increased by 21,3%.
AI education is expanding. Two-thirds of countries have already implemented or plan to implement The K–12 Computer Science Framework in schools (double the number in 2019). In the US, 81% of computer science teachers believe AI should be covered, but less than half are willing to teach it.
Industry leads in model development, but competition is growing. In 2024, almost 90% of all significant models were created in industry (in 2023 — only 60%). The computing power for training models doubles every 5 months. But the gaps between the leaders are narrowing: the difference between the best and tenth models fell from 11,9% to 5,4%, and between the first and second — only 0,7%.
AI has received the highest honors for its impact on science. Two Nobel Prizes have been awarded for research related to deep learning (physics) and its application to protein structure prediction (chemistry). The Turing Award is for achievements in reinforcement learning.
Complex reasoning is still a weak point for AI. While AI does well on Olympiad math problems, models still fail logic problems with known correct solutions. This limits their effectiveness in critical areas where absolute accuracy is required.
